Benchmark Almanac

Elimination Game

GPT-5.2 8 leads since Dec 2025 Not in model cards

Models play a multi-player game of public and private talk, alliances and votes to eliminate each other. A jury of eliminated players picks the winner.

Tests social reasoning, strategy and deception. About 300 GitHub stars.

Made by Lech Mazur Released Feb 2025 Multi-agent games All tasks public Last model added Dec 2025 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Ranked by matches Models play against each other, and the results become ratings.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
lechmazur/elimination_game 298 stars Leaderboard
57
Entries on its own leaderboard, 56 models we track

GPT-5.2 leads with 8 among models released by Dec 2025

Best score per entry from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

The newest model with a score here came out in Dec 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

57 entries. Scroll sideways for the rest.

  1. 8 GPT-5.2
  2. 6 GPT-5
  3. 6 GPT-5 mini
  4. 6 Claude Opus 4.5
  5. 6 Gemini 3 Flash
  6. 6 Grok-3 mini
  7. 5 GPT-4o
  8. 5 DeepSeek-R1
  9. 5 Claude 3.7 Sonnet
  10. 5 Claude Opus 4.1
  11. 5 Claude Sonnet 4.5
  12. 5 Grok 4
  13. 5 GPT-4.5
  14. 5 Claude 3.5 Sonnet (October 2024)
  15. 5 Grok 3
  16. 5 Gemini 3 Pro
  17. 5 Gemini 2.5 Flash
  18. 5 Claude Sonnet 4
  19. 5 MiniMax-M2
  20. 4 Qwen3-Max
  21. 4 o3
  22. 4 Gemini 2.5 Pro
  23. 4 Claude Opus 4
  24. 4 Qwen3-235B-A22B-Instruct
  25. 4 o3-mini
  26. 4 Kimi K2 Thinking
  27. 4 GLM-4.5
  28. 4 Mistral Large 2
  29. 4 DeepSeek-V3
  30. 4 DeepSeek-R1
  31. 4 o1
  32. 4 gpt-oss-120b
  33. 4 Mistral Large 3
  34. 4 Llama 4 Maverick
  35. 4 Grok 4.1 Fast
  36. 4 Llama 3.3 70B
  37. 4 Amazon Nova Pro
  38. 4 Qwen3-235B-A22B
  39. 4 GPT-4o Feb 2025
  40. 3 MiniMax-Text-01
  41. 3 Kimi K2
  42. 3 Mistral Small 3
  43. 3 Grok-2
  44. 3 GPT-4o mini
  45. 3 o4-mini
  46. 3 Claude 3.5 Haiku
  47. 3 Llama 3.1-405B
  48. 3 Gemini 2.0 Pro
  49. 3 Phi-4
  50. 3 Gemini 2.0 Flash Thinking
  51. 3 GLM-4.6
  52. 2 Qwen2.5-Max
  53. 2 Mistral Medium 3
  54. 2 QwQ-32B
  55. 2 Gemini 2.0 Flash
  56. 2 Qwen3-30B-A3B
  57. 0 Mistral Medium 3.1
  1. 1 GPT-5.2 medium 8
  2. 2 GPT-5 medium 6
  3. 3 GPT-5 mini medium 6
  4. 4 Claude Opus 4.5 thinking 16K 6
  5. 5 Gemini 3 Flash 6
  6. 6 Grok-3 mini high 6
  7. 7 GPT-4o Mar 2025 5
  8. 8 DeepSeek-R1 May 2025 5
  9. 9 Claude 3.7 Sonnet thinking 16K 5
  10. 10 Claude Opus 4.1 no reasoning 5
  11. 11 Claude Sonnet 4.5 thinking 16K 5
  12. 12 Grok 4 5
  13. 13 GPT-4.5 5
  14. 14 Claude 3.5 Sonnet (October 2024) 5
  15. 15 Grok 3 no reasoning 5
  16. 16 Gemini 3 Pro 5
  17. 17 Gemini 2.5 Flash Jun 2025 5
  18. 18 Claude Sonnet 4 no reasoning 5
  19. 19 MiniMax-M2 5
  20. 20 Qwen3-Max 4
  21. 21 o3 medium 4
  22. 22 Gemini 2.5 Pro Jun 2025 4
  23. 23 Claude Opus 4 no reasoning 4
  24. 24 Qwen3-235B-A22B-Instruct Jul 2025 4
  25. 25 o3-mini medium 4
  26. 26 Kimi K2 Thinking 4
  27. 27 GLM-4.5 4
  28. 28 Mistral Large 2 Nov 2024 4
  29. 29 DeepSeek-V3 4
  30. 30 DeepSeek-R1 4
  31. 31 o1 medium 4
  32. 32 gpt-oss-120b 4
  33. 33 Mistral Large 3 4
  34. 34 Llama 4 Maverick 4
  35. 35 Grok 4.1 Fast reasoning 4
  36. 36 Llama 3.3 70B 4
  37. 37 Amazon Nova Pro 4
  38. 38 Qwen3-235B-A22B 4
  39. 39 GPT-4o Feb 2025 4
  40. 40 MiniMax-Text-01 3
  41. 41 Kimi K2 Jul 2025 3
  42. 42 Mistral Small 3 3
  43. 43 Grok-2 Dec 2024 3
  44. 44 GPT-4o mini 3
  45. 45 o4-mini high 3
  46. 46 Claude 3.5 Haiku 3
  47. 47 Llama 3.1-405B 3
  48. 48 Gemini 2.0 Pro 3
  49. 49 Phi-4 3
  50. 50 Gemini 2.0 Flash Thinking Jan 2025 3
  51. 51 GLM-4.6 3
  52. 52 Qwen2.5-Max 2
  53. 53 Mistral Medium 3 2
  54. 54 QwQ-32B 2
  55. 55 Gemini 2.0 Flash Feb 2025 2
  56. 56 Qwen3-30B-A3B 2
  57. 57 Mistral Medium 3.1 0
#EntryScoreSettingDateSource
1 GPT-5.2 8 medium Dec 2025 Official leaderboard
2 GPT-5 6 medium Aug 2025 Official leaderboard
3 GPT-5 mini 6 medium Aug 2025 Official leaderboard
4 Claude Opus 4.5 6 thinking 16K Nov 2025 Official leaderboard
5 Gemini 3 Flash 6 Dec 2025 Official leaderboard
6 Grok-3 mini 6 high Apr 2025 Official leaderboard
7 GPT-4o Mar 2025 5 Mar 2025 Official leaderboard
8 DeepSeek-R1 May 2025 5 May 2025 Official leaderboard
9 Claude 3.7 Sonnet 5 thinking 16K Feb 2025 Official leaderboard
10 Claude Opus 4.1 5 no reasoning Aug 2025 Official leaderboard
11 Claude Sonnet 4.5 5 thinking 16K Sep 2025 Official leaderboard
12 Grok 4 5 Jul 2025 Official leaderboard
13 GPT-4.5 5 Feb 2025 Official leaderboard
14 Claude 3.5 Sonnet (October 2024) 5 Oct 2024 Official leaderboard
15 Grok 3 5 no reasoning Apr 2025 Official leaderboard
16 Gemini 3 Pro 5 Nov 2025 Official leaderboard
17 Gemini 2.5 Flash Jun 2025 5 Jun 2025 Official leaderboard
18 Claude Sonnet 4 5 no reasoning May 2025 Official leaderboard
19 MiniMax-M2 5 Oct 2025 Official leaderboard
20 Qwen3-Max 4 Sep 2025 Official leaderboard
21 o3 4 medium Apr 2025 Official leaderboard
22 Gemini 2.5 Pro Jun 2025 4 Jun 2025 Official leaderboard
23 Claude Opus 4 4 no reasoning May 2025 Official leaderboard
24 Qwen3-235B-A22B-Instruct Jul 2025 4 Jul 2025 Official leaderboard
25 o3-mini 4 medium Jan 2025 Official leaderboard
26 Kimi K2 Thinking 4 Nov 2025 Official leaderboard
27 GLM-4.5 4 Aug 2025 Official leaderboard
28 Mistral Large 2 Nov 2024 4 Nov 2024 Official leaderboard
29 DeepSeek-V3 4 Dec 2024 Official leaderboard
30 DeepSeek-R1 4 Jan 2025 Official leaderboard
31 o1 4 medium Dec 2024 Official leaderboard
32 gpt-oss-120b 4 Aug 2025 Official leaderboard
33 Mistral Large 3 4 Dec 2025 Official leaderboard
34 Llama 4 Maverick 4 Apr 2025 Official leaderboard
35 Grok 4.1 Fast 4 reasoning Nov 2025 Official leaderboard
36 Llama 3.3 70B 4 Dec 2024 Official leaderboard
37 Amazon Nova Pro 4 Dec 2024 Official leaderboard
38 Qwen3-235B-A22B 4 Apr 2025 Official leaderboard
39 GPT-4o Feb 2025 4 Official leaderboard
40 MiniMax-Text-01 3 Jan 2025 Official leaderboard
41 Kimi K2 Jul 2025 3 Jul 2025 Official leaderboard
42 Mistral Small 3 3 Jan 2025 Official leaderboard
43 Grok-2 Dec 2024 3 Dec 2024 Official leaderboard
44 GPT-4o mini 3 Jul 2024 Official leaderboard
45 o4-mini 3 high Apr 2025 Official leaderboard
46 Claude 3.5 Haiku 3 Oct 2024 Official leaderboard
47 Llama 3.1-405B 3 Jul 2024 Official leaderboard
48 Gemini 2.0 Pro 3 Feb 2025 Official leaderboard
49 Phi-4 3 Dec 2024 Official leaderboard
50 Gemini 2.0 Flash Thinking Jan 2025 3 Jan 2025 Official leaderboard
51 GLM-4.6 3 Sep 2025 Official leaderboard
52 Qwen2.5-Max 2 Jan 2025 Official leaderboard
53 Mistral Medium 3 2 May 2025 Official leaderboard
54 QwQ-32B 2 Mar 2025 Official leaderboard
55 Gemini 2.0 Flash Feb 2025 2 Feb 2025 Official leaderboard
56 Qwen3-30B-A3B 2 Apr 2025 Official leaderboard
57 Mistral Medium 3.1 0 Aug 2025 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Claude 3.7 Sonnet
thinking 16K
5
Claude Sonnet 4
no reasoning thinking 16K
5
Claude Opus 4
no reasoning thinking 16K
4
ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.2 medium 8
GPT-5 medium 6
GPT-5 mini medium 6
Claude Opus 4.5 thinking 16K 6
Gemini 3 Flash 6
Grok-3 mini high 6
GPT-4o Mar 2025 5
DeepSeek-R1 May 2025 5
Claude 3.7 Sonnet thinking 16K 5
Claude Opus 4.1 no reasoning 5
Claude Sonnet 4.5 thinking 16K 5
Grok 4 5
GPT-4.5 5
Claude 3.5 Sonnet (October 2024) 5
Grok 3 no reasoning 5
Gemini 3 Pro 5
Claude 3.7 Sonnet 5
Gemini 2.5 Flash Jun 2025 5
Claude Sonnet 4 no reasoning 5
MiniMax-M2 5
Qwen3-Max 4
o3 medium 4
Gemini 2.5 Pro Jun 2025 4
Claude Opus 4 no reasoning 4
Qwen3-235B-A22B-Instruct Jul 2025 4
o3-mini medium 4
Kimi K2 Thinking 4
Claude Sonnet 4 thinking 16K 4
GLM-4.5 4
Mistral Large 2 Nov 2024 4
DeepSeek-V3 4
DeepSeek-R1 4
Claude Opus 4 thinking 16K 4
o1 medium 4
gpt-oss-120b 4
Gemini 2.5 Pro Jun 2025 4
Mistral Large 3 4
Llama 4 Maverick 4
Grok 4.1 Fast reasoning 4
Llama 3.3 70B 4
Amazon Nova Pro 4
Qwen3-235B-A22B 4
MiniMax-Text-01 3
Kimi K2 Jul 2025 3
Mistral Small 3 3
Grok-2 Dec 2024 3
GPT-4o mini 3
o4-mini high 3
Claude 3.5 Haiku 3
Llama 3.1-405B 3
Gemini 2.0 Pro 3
Phi-4 3
Gemini 2.0 Flash Thinking Jan 2025 3
GLM-4.6 3
Qwen2.5-Max 2
Mistral Medium 3 2
QwQ-32B 2
Gemini 2.0 Flash Feb 2025 2
Qwen3-30B-A3B 2
Mistral Medium 3.1 0

6 models have held number one since its release in Feb 2025

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic OpenAI xAI
Own leaderboard Not tested since Dec 2025
GPT-5.2 8

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard
0 2 4 6 8 2025 2026 0 2 4 6 8 2025 2026

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Games and long tasks

All 68 in Games and long tasks
  • ARC-AGI-3 In use Interactive game environments where agents must explore, infer goals and rules, and act efficiently without instructions. GPT-6 Astra 100%
  • MineDojo Fading Agents complete open-ended tasks in Minecraft, such as mining blocks and fighting mobs, from a large simulation suite. Programmatic checks score many tasks. Seed2.0 Pro 49%
  • Orak Fading LLM agents play 12 popular video games across most genres through an MCP interface. Scores come from in-game progress or outcomes per title. Nemotron 3 Nano 4B 22.9
  • GameWorld Only in system cards Multimodal game agents play 34 browser games across 170 tasks. Outcomes are checked from the game state. Gemini 3 Flash 41.9%
  • Swarm Bench Only in system cards Models orchestrate agent swarms on complex tasks that benefit from coordinated decomposition and parallel execution. Kimi K3 76.3%
  • TextQuests Only in system cards Agents play 25 classic interactive fiction games. Measures game progress and harmful actions per 100 steps. Muse Spark 2.40